Generative AI has raised some of the most contested copyright questions in decades. Authors, artists, musicians, news organizations and software developers have filed lawsuits against AI companies, while those companies argue that their use of data is lawful. Meanwhile, anyone using AI to create content wants to know whether they own the result.
This is an overview, not legal advice, and the law is developing quickly and differs between countries.
Updated September 2026: we added links to the court decisions, official reports and legislation, and brought the article up to date with the final approval of the Anthropic settlement and the UK government’s March 2026 report.
Question 1: Is training on copyrighted material allowed?
AI models are trained on vast amounts of text, images, audio and code, much of it copyrighted. The central dispute is whether that training requires permission.
The AI companies’ argument, in the US, is typically that training is fair use: the models learn patterns rather than storing copies, and the use is transformative.
Rights holders argue that their work was copied without permission or payment to build commercial products that can compete with them, and that models can sometimes reproduce protected material.
What US courts have said so far
| Case | Ruling | Status |
|---|---|---|
| Bartz v. Anthropic (N.D. Cal., June 2025) | Training on lawfully acquired books was fair use, described as “spectacularly” transformative. Keeping pirated copies in a central library was not protected. | Settled for $1.5 billion; final approval July 2026 |
| Kadrey v. Meta (N.D. Cal., June 2025) | Meta won on the evidence presented, but the judge said AI training could harm authors by flooding the market with competing works. | Ruling limited to these plaintiffs’ arguments |
| Thomson Reuters v. Ross Intelligence (D. Del., February 2025) | Using Westlaw headnotes to train a competing legal research tool was not fair use. The AI involved was not generative. | Appeal argued at the Third Circuit on 11 June 2026; no ruling published as of late September 2026 |
Sources: the Ohio State University Libraries 2026 fair use update, the Authors Guild and LawSites. These early decisions reach different conclusions on different facts, including how the training data was obtained, and appeals will shape the law. The Third Circuit’s decision in the Ross case is likely to be the first appellate ruling on AI training.
The Anthropic settlement. In the Bartz case, Anthropic agreed to pay $1.5 billion, about $3,000 per book, to settle claims over books downloaded from pirate sites. A federal court in California granted final approval on 20 July 2026. The settlement covers only past conduct involving those pirated copies, and authors kept the right to bring claims about AI outputs.
Other jurisdictions take different approaches. The EU’s Copyright in the Digital Single Market Directive allows text and data mining unless rights holders opt out in a machine-readable way, and the EU AI Act requires general-purpose model providers to have copyright policies and publish summaries of training content. In the UK, the government had favored a similar opt-out, but its March 2026 report on copyright and AI dropped it as the preferred option after consultation responses highlighted the practical and economic difficulties of opt-out schemes.
Question 2: Can AI output infringe copyright?
Yes, it can. If an AI tool produces something substantially similar to an existing protected work, publishing or selling it could infringe, regardless of whether a machine produced it. Risks are higher when prompts ask for content “in the style of” specific works or mention characters and brands. Most AI companies add safeguards against reproducing protected material, and some offer business customers legal indemnity for outputs.
Question 3: Who owns AI-generated content?
In the United States, the Copyright Office’s January 2025 report on copyrightability concluded that “prompts do not alone provide sufficient control” for the result to have a human author. Material generated entirely by AI from a prompt is therefore generally not protected. Human authors can still claim copyright in their own expression that is visible in the output, in the creative selection and arrangement of material, and in creative changes they make to it. The Office has published its full series of AI reports, including a report on training.
Other countries’ rules differ. The UK is unusual in protecting “computer-generated works” with no human author, but the government said in March 2026 that it proposes to remove that protection, according to Fieldfisher‘s summary of the report.
Practical implication If owning the rights to your content matters, for example a logo, a book or a product design, make sure there is meaningful human creative contribution, and keep records of your process.
Licensing is growing
Alongside the lawsuits, many publishers, stock image libraries and other rights holders have signed licensing deals with AI companies, providing content in exchange for payment and, sometimes, attribution. This market is likely to keep growing whatever the courts decide.
What creators can do
- Review the terms of platforms where you publish work.
- Use available opt-out mechanisms, such as controls in robots.txt for AI crawlers, knowing they are not universally respected. In the EU, a machine-readable opt-out has legal effect under the text and data mining rules.
- Consider collective organizations negotiating on creators’ behalf.
What businesses using AI should do
- Prefer tools with clear commercial terms and, where available, indemnities.
- Avoid prompts that target specific copyrighted works, characters or living artists’ styles.
- Keep humans involved in creating content you need to own.
- Get legal advice for high-value uses.
A law still being written
The copyright rules for AI are being written right now, in courtrooms and legislatures. Early US rulings lean toward treating training on lawfully obtained material as fair use, while copying from pirate sites has proved costly, and appeals are pending. Until the law settles, the safest path is to use reputable tools, add genuine human creativity and treat outputs with the same care as any other content you publish.
Sources
- Fair Use and Artificial Intelligence: 2026 Update, Ohio State University Libraries, March 2026
- Court Grants Final Approval of $1.5 Billion Anthropic Copyright Settlement, The Authors Guild
- At 3rd Circuit, Judges Press ROSS and Thomson Reuters on Fair Use, LawSites, June 2026
- Copyright and Artificial Intelligence, Part 2: Copyrightability, US Copyright Office, January 2025
- Copyright and Artificial Intelligence (all reports), US Copyright Office
- Directive (EU) 2019/790 on copyright in the Digital Single Market, EUR-Lex
- Report on Copyright and Artificial Intelligence, UK Government, March 2026
- UK Government maintains status quo on AI and copyright, Fieldfisher
